CN109344751A - A kind of reconstructing method of internal car noise signal - Google Patents

A kind of reconstructing method of internal car noise signal Download PDF

Info

Publication number
CN109344751A
CN109344751A CN201811102322.1A CN201811102322A CN109344751A CN 109344751 A CN109344751 A CN 109344751A CN 201811102322 A CN201811102322 A CN 201811102322A CN 109344751 A CN109344751 A CN 109344751A
Authority
CN
China
Prior art keywords
component
signal
value
fitness value
internal car
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201811102322.1A
Other languages
Chinese (zh)
Other versions
CN109344751B (en
Inventor
王孝兰
杨东坡
王岩松
郭辉
刘宁宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai University of Engineering Science
Original Assignee
Shanghai University of Engineering Science
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai University of Engineering Science filed Critical Shanghai University of Engineering Science
Priority to CN201811102322.1A priority Critical patent/CN109344751B/en
Publication of CN109344751A publication Critical patent/CN109344751A/en
Application granted granted Critical
Publication of CN109344751B publication Critical patent/CN109344751B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
  • Fittings On The Vehicle Exterior For Carrying Loads, And Devices For Holding Or Mounting Articles (AREA)

Abstract

The present invention relates to a kind of reconstructing methods of internal car noise signal, comprising the following steps: 1) signal decomposition is analyzed: carrying out signal decomposition to source signal and analyzes to obtain three stable signal component classifications, i.e. high fdrequency component, intermediate frequency component and low frequency component；2) component fitness calculates: building BP neural network model is trained respectively, and using the weight of BP neural network model and threshold value as component fitness value, obtains optimal component fitness value；3) signal reconstruction model: according to the classification of input signal component, optimal component fitness value is assigned to noise reconstruct BP network as initial weight and threshold value, it is trained, the corresponding restructing algorithm model of every class signal component is obtained after convergence, and the reconstruct that occupant's ear side noise signal is completed in superposition is reconstructed according to restructing algorithm model.Compared with prior art, the present invention has many advantages, such as to reduce the non-stationary of signal and modeling difficulty, improves reconstruction accuracy.

Description

A kind of reconstructing method of internal car noise signal
Technical field
The present invention relates to signal processings and information to merge field, more particularly, to a kind of reconstruct side of internal car noise signal Method.
Background technique
In order to realize the active control (ANC) of passenger's ear side noise, first have to provide primary reference for control system Signal.Pickup for primary reference signal, conventional method are to install microphone near occupant's ear side to obtain primary reference Signal, the method inevitably introduce secondary sound source secondary pollution, are unfavorable for the fast convergence of system.Therefore, vehicle is studied The reconstructing method of interior occupant's ear side noise, the reference signal for obtaining ANC have certain meaning.
Currently, the main method of Reconstruction of Sound Field includes near field acoustic holography (NAH) and Fusion (MSDF) etc.. Time domain NAH method needs to establish on the basis of free found field is assumed, but is generally difficult to meet in practical application.In view of occupant Noise reconstruct in ear side utilizes pass-by noise source signal to realize, reconstruct obtains occupant's ear side signal.Data fusion method goes out Now theoretical foundation is provided to solve multi-source data feature extraction and integration modeling.Multisource data fusion passes through established rule Then and analysis method, it from multi-sensor data and informix, and is obtained on this basis to the consistency of target being observed Understanding.
MSDF method is passed from Kalman filtering method, Bayes' assessment, maximum likelihood estimate, clustering methodology etc. System method identifies (FLI), support vector machines to based on Self-organizing Maps (SOM), adaptive weighted fusion (AWF), fuzzy logic (SVM) develop with the intelligent direction of artificial neural network (ANN).Wherein ANN is widely used, and wherein most is (anti-using BP To propagation) algorithm.BP neural network establishes model according to the inner link of data itself, extracts correlation from data automatically and knows Know, and there is self study, self-organizing and adaptive ability.
However, internal car noise signal belongs to mechanical oscillation and acoustic signal, there is strong nonlinearity and non-stationary, although BP Neural network has good nonlinear fitting ability, can be advantageously applied to multisource data fusion technology, but processing has The signal of non-stationary feature, bring error are inevitable.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of internal car noise signals Reconstructing method.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of reconstructing method of internal car noise signal, comprising the following steps:
1) signal decomposition is analyzed: being obtained normal noise source signal data and is pre-processed, and carries out letter to source signal Number decomposition analysis obtains three stable signal component classifications, i.e. high fdrequency component, intermediate frequency component and low frequency component；
2) component fitness calculates: respectively using the noise source signal component of three classifications as input, being believed with internal car noise Number component constructs BP neural network model and is trained respectively as desired output, and by the weight of BP neural network model and Threshold value obtains optimal component fitness value as component fitness value；
3) signal reconstruction model: according to the classification of input signal component, optimal component fitness value is assigned to noise reconstruct BP network is trained as initial weight and threshold value by error back propagation method, obtains every class signal component after convergence The reconstruct that occupant's ear side noise signal is completed in superposition is reconstructed according to restructing algorithm model in corresponding restructing algorithm model.
The step 1) specifically includes the following steps:
11) source signal X (t) is decomposed based on intrinsic mode decomposition, then had:
Wherein, ciIt (t) is i-th of IMF component of signal, n is IMF component sum, rnFor the residual components of output signal；
12) each IMF component is obtained to account in interior extreme value points per second and each IMF component in the energy of original signal Than；
13) extreme point number threshold value (N, M) is set, IMF component is divided by high frequency according to the extreme point number of each component Component, intermediate frequency component and low frequency component, then have:
X (t)=d1+d2+d3
Wherein, d1,d2,d3Respectively indicate high fdrequency component, intermediate frequency component and low frequency component.
In the step 2), the BP neural network model of building is one or three layers of BP network.
In the step 2), the expression formula of component fitness value are as follows:
Wherein,For diThe corresponding component fitness value of component, and i=1,2,3,For the weight of input layer to hidden layer Matrix,For the weight matrix of hidden layer to output layer,For hidden layer threshold value,For output layer threshold value.
In the step 2), when the output learning error of BP neural network model be less than desired output error when or iteration When step number reaches maximum setting step number, corresponding component fitness value is optimal component fitness value at this time, then has:
Wherein,For optimal component fitness value,The component fitness value updated for iteration r step.
In the step 13), extreme point number threshold value (N, the M) value is (80,20).
Compared with prior art, the invention has the following advantages that
The reconstruct of cab signal may be implemented in signal reconfiguring method through the invention, by SDA method, from largely On reduce the non-stationary of signal, reduce modeling difficulty；Adaptive optimal control angle value is obtained by CFC simultaneously and assigns noise reconstruct BP Network can solve BP network weight and problem be randomly generated in threshold value as initial weight and threshold value, and then improve reconstruction accuracy
Detailed description of the invention
Fig. 1 is the flow diagram of signal reconfiguring method of the present invention.
Fig. 2 is the neural network structure figure of noise reconstruct.
Fig. 3 is the contribution amount analysis chart of each key point.
Fig. 4 is the energy accounting figure of each IMF component.
Fig. 5 (a) is that set forth herein the signal reconstructions of algorithm and original signal result in time domain comparison diagram.
Fig. 5 (b) is that set forth herein the signal reconstructions of algorithm and original signal frequency-domain result comparison diagram.
Fig. 6 (a) is the signal reconstruction and original signal result in time domain comparison diagram of BP restructing algorithm.
Fig. 6 (b) is the signal reconstruction and original signal frequency-domain result comparison diagram of BP restructing algorithm.
Specific embodiment
Based on the above the deficiencies in the prior art, technical problem solved by the invention is to provide a kind of cab signal reconstruct Method, in order to reconstruct internal car noise signal, to influence the key point signal of internal car noise as input data, based on BP nerve Network establishes noise reconstruction model, obtains occupant's ear side noise signal.In view of high-speed working condition noise signal feature, BP nerve net The limitation of network, firstly, SDA method rebuilds signal decomposition, reduction original signal is non-stationary, obtains three opposite smoothness Higher signal component, and instruct different components according to the movement of respective classification based on neural network model, and define CFC and come more The fitness value of new component, wherein fitness value indicates the weight and threshold value of neural network.Then, a large amount of three classes signals point are utilized Amount data are trained it, obtain optimal component fitness value, the weight and threshold value to network carry out initialization assignment.Most Afterwards, the internal car noise signal intelligent restructing algorithm model based on BP neural network is obtained using test data.By verification experimental verification, Algorithm is can effectively to reconstruct occupant's ear side noise signal.
To achieve the above object, the present invention adopts the following technical scheme:
A kind of internal car noise signal reconfiguring method based on source signal, comprising the following steps:
Step 1 obtains normal noise source signal data, determines the crucial noise source signal for influencing internal car noise signal, Data are pre-processed；
In step 1:
In semianechoic room, sound-acoustic transfer function of each transmission path is obtained based on reciprocity method, is guaranteed test position and is obtained Road test position consistency is taken, the crucial measuring point signal of most correlation is selected to reconstruct internal car noise；
Analysis method is contributed using acoustic transfer function (TPA), calculates tribute of each noise source signal to cab signal The amount of offering, as shown in formula:
Wherein, PMNFor total acoustic pressure of passenger's ear side；For the pressure contribution amount on transmission path i；For road Transmission characteristic function on diameter i；For the work input on transmission path i.
Step 2, data-signal progress SDA method are decomposed and reconstituted, in step 2:
Data-signal passes through EMD resolution process, obtains limited metastable IMF component and a trend term rn；
Using EED method proposed in this paper, the extreme value points and energy accounting of each IMF component are solved, it is suitable to determine (N, M) value obtains 3 components, i.e. high fdrequency component, intermediate frequency component, low frequency component.
X (t)=d1+d2+d3 (3)
One step 3, design 3 layers of BP neural network structure, noise source signal are made as input signal, internal car noise signal For desired output；
Step 4, the weight threshold of BP network are randomly generated, and are carried out according to weight threshold to component fitness value Coding, in step 4:
I-th noise like source signal component is as input signal (i=1,2,3), and the i-th class internal car noise signal component is as the phase Hope output signal, initial fitness valueIt is randomly generated, the weight and threshold value of neural network are as component fitness value；
Using the three-layer neural network basic structure of n-m-l, wherein n, m, l respectively represent input layer, hidden layer and defeated Layer neuron number out, and the weight of network and threshold matrix are expressed as follows；
Initial component fitness valueIt can indicate are as follows:
Step 5, optimal fitness value calculation, in steps of 5:
The number of iterations initial value r=0；Compare minimum learning error and meets anticipation error；
The output signal O of the propagated forward of neuron j is calculated using input dataj；
Wherein, wijAnd θjThe weight and threshold value of neuron j are respectively indicated, f is excitation function, and excitation function is using hyperbolic just Cut sigmoid function.
After the r times iteration, i-th of component fitness value is updated
Calculate error E；
Wherein yd,kAnd ykThe desired output and true output of output layer neuron are respectively represented, N is data sample Number.
If E≤γ or r >=max_times, output signal component diAdaptive optimal control angle value
Step 6, internal car noise signal reconstruction model foundation, in step 6:
Noisy test signal is decomposed and is rebuild, process step 1；
Judge signal component diWith optimal appropriateness valueBelong to same class,By matrix 4,5,6,7, assign Noise reconstructs BP network as initial weight and threshold value, and utilizes signal component diTraining noise reconstructs BP network；
The number of iterations initial value r=0；Compare minimum learning error and meets anticipation error；
The output signal O of the propagated forward of neuron j is calculated using input dataj；
Error E is calculated, if E≤γ or r >=max_times, obtains signal component diNoise reconstruct BP network.
Above-mentioned process is repeated, entire signal reconstruction model is obtained；
Step 7, each noise source signal input signal reconstruction model export internal car noise reconstruction signal.
The embodiment of the invention will now be described in detail with reference to the accompanying drawings, and as part of this specification passes through Embodiment come with illustrate the principle of the present invention.
Embodiment
Below with the normal noise source signal data under the high-speed working condition of acquisition, as shown in figure 3, the tribute of each key point The amount of offering analysis chart determines the crucial noise source signal for influencing internal car noise signal.Data are pre-processed, and noise source signal is obtained Sample database determines that the input layer number of BP network is n=4, output layer l=1.
As shown in Figure 1, be the flow diagram of signal reconstruction of the embodiment of the present invention,
The first step, SDA method carry out decomposition reconstruction to data-signal, and data-signal passes through EMD resolution process, obtains limited A metastable IMF component and a trend term rn；
Using EMD methods proposed in this paper, the extreme value points and energy accounting of each IMF component are solved, see figure and table 1, Determine that suitable (N, M)=(80,20) obtain 3 components, i.e. high fdrequency component, intermediate frequency component, low frequency component.
X (t)=d1+d2+d3 (2)
The extreme point number (/s) and energy ratio of 1 internal car noise signal IMF component of table
Second step determines 3 layers of BP neural network basic structure using 4-m-1, is based on this signal source m=128, such as Shown in Fig. 2, noise source signal is as input signal, and internal car noise signal is as desired output；
The weight and threshold matrix of network are expressed as follows；
The weight and threshold value of neural network are as component fitness value, Initial component fitness valueIt can indicate are as follows:
Third step, optimal fitness value calculation:
I-th noise like source signal component is as input signal (i=1,2,3), and the i-th class internal car noise signal component is as the phase Hope output signal, initial fitness valueIt is randomly generated,；
The number of iterations initial value r=0；Compare minimum learning error and meets anticipation error；
The output signal O of the propagated forward of neuron j is calculated using input dataj；
Wherein, wijAnd θjThe weight and threshold value of neuron j are respectively indicated, f is excitation function, and excitation function is using hyperbolic just Cut sigmoid function.
After the r times iteration, i-th of component fitness value is updated
Calculate error E；
Wherein yd,kAnd ykThe desired output and true output of output layer neuron are respectively represented, N is data sample Number.
If E≤γ or r >=max_times, output signal component diAdaptive optimal control angle value
4th step establishes signal reconstruction model:
Noisy test signal is decomposed and is rebuild；
Judge signal component diWith optimal appropriateness valueBelong to same class,By matrix 3,4,5,6, assign Noise reconstructs BP network as initial weight and threshold value, and utilizes signal component diTraining noise reconstructs BP network；
The number of iterations initial value r=0；Compare minimum learning error and meets anticipation error；
The output signal O of the propagated forward of neuron j is calculated using input dataj；
Error E is calculated, if E≤γ or r >=max_times, obtains signal component diNoise reconstruct BP network.
It repeats above-mentioned process and obtains entire signal reconstruction model.
Validation verification is carried out to internal car noise signal reconstruction algorithm model of the present invention.
Fig. 5 indicates the comparison of the result and original signal of signal reconstruction.It can be seen from the figure that the nerve of the proposition of this paper Network algorithm can correctly reconstruct position and the amplitude of original signal to internal car noise signal, the results showed that the method have compared with High signal reconstruction performance.
Algorithms of different carries out performance comparison
Fig. 5 and Fig. 6 carries out performance comparison to signal reconstruction BP network and network algorithm model proposed in this paper.It can from figure To find out, two kinds of restructing algorithms can correctly reconstruct position and the amplitude of original signal to internal car noise signal, the results showed that The method signal reconstruction performance with higher.
In order to further analyze reconstruction result precision, characterized herein using mean square error E, shown in formula.
Wherein yd,iAnd yiReconstruction value and true value are respectively indicated, N is sample number.
In order to preferably make an explanation to result, to yd,i、yiIt is normalized.Shown in formula.
Wherein, ymin,ymaxRespectively represent minimum value, the maximum value of vector y.y1Indicate normalized result.
It normalizes result to substitute into formula (11), the root mean square relative error reconstructed such as table 2.
The root mean square relative error that table 2 reconstructs
From Table 2, it can be seen that for occupant's ear side noise signal reconstruction under high-speed working condition, algorithm proposed in this paper Model is better than BP algorithm model, wherein the result precision not being normalized improves 39.33%, normalized result precision 68.92% is improved, wherein normalization resultant error is 0.0023 less than anticipation error 0.005, this illustrates proposed in this paper Internal car noise signal reconstruction model can satisfy required precision
To sum up, the reconstruct of internal car noise signal through the invention, can be to avoid the secondary pollution of secondary sound source.Lead to simultaneously Crossing signal reconfiguring method of the invention may be implemented the reconstruct of cab signal, by SDA method, from largely reducing letter Number it is non-stationary, reduce modeling difficulty；Adaptive optimal control angle value is obtained by CFC simultaneously and assigns noise reconstruct BP network as just Beginning weight and threshold value, can solve BP network weight and problem is randomly generated in threshold value, and then improve reconstruction accuracy.

Claims (6)

1. a kind of reconstructing method of internal car noise signal, which comprises the following steps:
1) signal decomposition is analyzed: being obtained normal noise source signal data and is pre-processed, and carries out signal point to source signal Solution analysis obtains three stable signal component classifications, i.e. high fdrequency component, intermediate frequency component and low frequency component；
2) component fitness calculates: respectively using the noise source signal component of three classifications as input, with internal car noise signal point Amount is used as desired output, and building BP neural network model is trained respectively, and by the weight and threshold value of BP neural network model As component fitness value, optimal component fitness value is obtained；
3) it signal reconstruction model: according to the classification of input signal component, assigns optimal component fitness value to noise and reconstructs BP net Network is trained as initial weight and threshold value by error back propagation method, and it is corresponding to obtain every class signal component after convergence Restructing algorithm model, according to restructing algorithm model be reconstructed superposition complete occupant's ear side noise signal reconstruct.
2. a kind of reconstructing method of internal car noise signal according to claim 1, which is characterized in that the step 1) tool Body the following steps are included:
11) source signal X (t) is decomposed based on intrinsic mode decomposition, then had:
Wherein, ciIt (t) is i-th of IMF component of signal, n is IMF component sum, rnFor the residual components of output signal；
12) obtain each IMF component interior extreme value points per second and each IMF component original signal energy accounting；
13) set extreme point number threshold value (N, M), according to the extreme point number of each component by IMF component be divided into high fdrequency component, Intermediate frequency component and low frequency component, then have:
X (t)=d1+d2+d3
Wherein, d1,d2,d3Respectively indicate high fdrequency component, intermediate frequency component and low frequency component.
3. a kind of reconstructing method of internal car noise signal according to claim 1, which is characterized in that the step 2) In, the BP neural network model of building is one or three layers of BP network.
4. a kind of reconstructing method of internal car noise signal according to claim 3, which is characterized in that the step 2) In, the expression formula of component fitness value are as follows:
Wherein,For diThe corresponding component fitness value of component, and i=1,2,3,For the weight matrix of input layer to hidden layer,For the weight matrix of hidden layer to output layer,For hidden layer threshold value,For output layer threshold value.
5. a kind of reconstructing method of internal car noise signal according to claim 4, which is characterized in that the step 2) In, when the output learning error of BP neural network model is less than desired output error or iterative steps reach maximum setting step number When, corresponding component fitness value is optimal component fitness value at this time, then have:
Wherein,For optimal component fitness value,The component fitness value updated for iteration r step.
6. a kind of reconstructing method of internal car noise signal according to claim 1, which is characterized in that the step 13) In, extreme point number threshold value (N, the M) value is (80,20).
CN201811102322.1A 2018-09-20 2018-09-20 Reconstruction method of noise signal in vehicle Active CN109344751B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811102322.1A CN109344751B (en) 2018-09-20 2018-09-20 Reconstruction method of noise signal in vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811102322.1A CN109344751B (en) 2018-09-20 2018-09-20 Reconstruction method of noise signal in vehicle

Publications (2)

Publication Number Publication Date
CN109344751A true CN109344751A (en) 2019-02-15
CN109344751B CN109344751B (en) 2021-10-08

Family

ID=65306251

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811102322.1A Active CN109344751B (en) 2018-09-20 2018-09-20 Reconstruction method of noise signal in vehicle

Country Status (1)

Country Link
CN (1) CN109344751B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111487063A (en) * 2020-03-20 2020-08-04 天津大学 Parameter adaptive internal combustion engine noise source separation method
CN111721401A (en) * 2020-06-17 2020-09-29 广州广电计量检测股份有限公司 Low-frequency noise analysis system and method
CN115035911A (en) * 2022-05-31 2022-09-09 中国第一汽车股份有限公司 Noise generation model training method, device, equipment and medium
CN117437931A (en) * 2023-12-20 2024-01-23 深圳市新厚泰电子科技有限公司 Sound signal optimized transmission method for microphone

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049798A (en) * 2012-12-05 2013-04-17 浙江大学城市学院 Short-period electric generation power forecasting method applied to photovoltaic electric generation system
CN103077267A (en) * 2012-12-28 2013-05-01 电子科技大学 Parameter sound source modeling method based on improved BP (Back Propagation) neural network
CN103474066A (en) * 2013-10-11 2013-12-25 福州大学 Ecological voice recognition method based on multiband signal reconstruction
CN104168131A (en) * 2014-06-05 2014-11-26 国家电网公司 Flow generation method of power dispatching exchange network based on multicast communication
CN106769040A (en) * 2016-12-14 2017-05-31 上海工程技术大学 A kind of method of the sparse reconstruct of bearing vibration signal

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049798A (en) * 2012-12-05 2013-04-17 浙江大学城市学院 Short-period electric generation power forecasting method applied to photovoltaic electric generation system
CN103077267A (en) * 2012-12-28 2013-05-01 电子科技大学 Parameter sound source modeling method based on improved BP (Back Propagation) neural network
CN103474066A (en) * 2013-10-11 2013-12-25 福州大学 Ecological voice recognition method based on multiband signal reconstruction
CN104168131A (en) * 2014-06-05 2014-11-26 国家电网公司 Flow generation method of power dispatching exchange network based on multicast communication
CN106769040A (en) * 2016-12-14 2017-05-31 上海工程技术大学 A kind of method of the sparse reconstruct of bearing vibration signal

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
YUE SHENG WANG等: "A study on the emulsifier fault diagnosis of the BP neural network based on EMD and KPCA", 《PROCEEDINGS OF THE 2016 2ND INTERNATIONAL CONFERENCE ON ENERGY EQUIPMENT SCIENCE AND ENGINEERING》 *
芮国胜等: " 一种基于基追踪压缩感知信号重构的改进算法", 《电子测量技术》 *
薛金亮: "基于振动噪声信息融合的齿轮箱齿轮故障诊断研究", 《中国优秀硕士学位论文全文数据库》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111487063A (en) * 2020-03-20 2020-08-04 天津大学 Parameter adaptive internal combustion engine noise source separation method
CN111487063B (en) * 2020-03-20 2021-05-14 天津大学 Parameter adaptive internal combustion engine noise source separation method
CN111721401A (en) * 2020-06-17 2020-09-29 广州广电计量检测股份有限公司 Low-frequency noise analysis system and method
CN111721401B (en) * 2020-06-17 2022-03-08 广州广电计量检测股份有限公司 Low-frequency noise analysis system and method
CN115035911A (en) * 2022-05-31 2022-09-09 中国第一汽车股份有限公司 Noise generation model training method, device, equipment and medium
CN117437931A (en) * 2023-12-20 2024-01-23 深圳市新厚泰电子科技有限公司 Sound signal optimized transmission method for microphone
CN117437931B (en) * 2023-12-20 2024-04-02 深圳市新厚泰电子科技有限公司 Sound signal optimized transmission method for microphone

Also Published As

Publication number Publication date
CN109344751B (en) 2021-10-08

Similar Documents

Publication Publication Date Title
CN109344751A (en) A kind of reconstructing method of internal car noise signal
CN111325112B (en) Cutter wear state monitoring method based on depth gate control circulation unit neural network
CN108766419B (en) Abnormal voice distinguishing method based on deep learning
CN107194404B (en) Underwater target feature extraction method based on convolutional neural network
CN109800700B (en) Underwater acoustic signal target classification and identification method based on deep learning
Weymaere et al. On the initialization and optimization of multilayer perceptrons
CN102201236B (en) Speaker recognition method combining Gaussian mixture model and quantum neural network
CN112435363B (en) Cutter wear state real-time monitoring method
CN110728360A (en) Micro-energy device energy identification method based on BP neural network
CN113571067B (en) Voiceprint recognition countermeasure sample generation method based on boundary attack
CN109272989A (en) Voice awakening method, device and computer readable storage medium
CN107085704A (en) Fast face expression recognition method based on ELM own coding algorithms
KR20160032536A (en) Signal process algorithm integrated deep neural network based speech recognition apparatus and optimization learning method thereof
CN110853656B (en) Audio tampering identification method based on improved neural network
CN110456332A (en) A kind of underwater sound signal Enhancement Method based on autocoder
Zeng et al. An end-to-end deep source recording device identification system for web media forensics
Asemi et al. Adaptive neuro-fuzzy inference system for evaluating dysarthric automatic speech recognition (ASR) systems: a case study on MVML-based ASR
CN107292852A (en) A kind of Image denoising algorithm theoretical based on low-rank
CN112270931A (en) Method for carrying out deceptive voice detection based on twin convolutional neural network
CN114663685B (en) Pedestrian re-recognition model training method, device and equipment
CN111816200B (en) Multi-channel speech enhancement method based on time-frequency domain binary mask
Lei et al. Towards recurrent neural network with multi-path feature fusion for signal modulation recognition
CN114067819B (en) Speech enhancement method based on cross-layer similarity knowledge distillation
Sharma et al. Novel approach to design matched digital filter with Abelian group and fuzzy particle swarm optimization vector quantization
Li et al. A Convolutional Neural Network with Non-Local Module for Speech Enhancement.

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant